Vercel Chat SDK now runs Claude Managed Agents, no backend
Ship a persistent per-thread chat agent with streamed traces and no state to run. Plus: two critical Next.js RCEs to patch and Cursor joins the AI SDK harness.
Copy markdown
One session per thread, zero database
The new Chat SDK + Claude Managed Agents template gives token-by-token streaming and one persistent session per chat thread, with conversation history held by the Managed Agents service — so there's no backend state to run. Deploy the sample Slack research bot in one click, or fork vercel-labs/cma-chat-sdk.
Stream the agent's trace into the chat
Tool calls and model requests are exposed during the turn, so you can render a live trace — the agent's steps and its sources — right in the UI as it works. The same template drops into Teams, Google Chat, Discord, WhatsApp and 30+ other platforms.
Patch Next.js now: two critical RCEs
Next.js's Aug 25 security release fixes two unauthenticated remote-code-execution bugs — one in AVIF image handling (via libheif), one on Windows-hosted servers. Self-hosters must upgrade to next@15.5.24 (15.x) or next@16.3.3 (16.x); apps on Vercel are already protected.
Cursor joins the AI SDK harness
Vercel's AI SDK harness layer now supports Cursor alongside Claude Code and Codex via @ai-sdk/harness-cursor, letting you spawn sandboxed agent sessions with AI SDK-compatible streams from inside your own app.
Elsewhere: filter deployments in plain English
Vercel's Deployments page revamps filtering with one-click suggestions, typeahead matching and natural-language queries, so you can find a build without memorizing the filter syntax.